Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
Welcome to Brainstuff from how stuff Works dot com where
smart Happens. Hi. I'm Marshall Brain with today's question, how
do self driving cars work? Who wouldn't like to have
a car that drives itself. The idea of being able
(00:22):
to sleep or to use your laptop during the commute
to work is certainly appealing. The idea of having your
car drop you off at the door of a store
and then go park itself is also attractive. And then
there's the safety thing. Human beings just are not that
good when it comes to driving, with millions of automobile
accidents causing damage, injury, and death every year. If we
(00:46):
all had self driving cars, it would be great. Two
recent announcements have made self driving cars seem a lot
closer to reality. The first comes from the Army, which
is now using self driving vehicle to guard a large
military facility and nuclear waste dump in Nevada. And then
there's Google, which recently announced that it's developed self driving
(01:09):
cars that have logged over a hundred thousand miles on
real roads in real traffic. What's going on inside these
cars to make it possible for them to drive themselves.
We can take a look at it. By looking at
Google cars or some of the cars that have participated
in DARPST Grand Challenge, you could, in theory, turn any
(01:31):
car into a self driving car. The first thing you
would have to do is make it possible for a
computer to control the car. This is not quite as
easy as it sounds, but it's relatively straightforward. The computer
needs to be able to turn the steering wheel, push
the accelerator and brake pedals, move the gear shift control,
and start the engine. These tasks, at least in experimental
(01:54):
self driving cars, are often accomplished with electric motors. A
motor is mounted so it can turn the existing steering
wheel for example. Another motor is able to put pressure
on the accelerator and brake pedals and so on. It's
not a particularly elegant solution, but it gets the job done.
As production cars become more advanced, drive by wire systems
(02:16):
are becoming a lot more common, So in a like
a Prius a Toyota Prius, if you push the brake pedal,
it might go to a computer and the computer actually
activates the brakes. There's no direct mechanical linkage between the
brake pedal and the brake mechanism that's called drive by
wire when the computer is sitting there in the middle.
(02:39):
In cars with drive by wire, a computer can hook
directly into that existing control system without having to have
these motors hanging out in the driver's area. The next
thing that a self driving car needs is sensors, and
here things get pretty complicated. Today, a human being drives
his car by using his or her eye is as
(03:00):
the soul sensor. The human visual system is amazing in
this regard because it's able to accurately judge the presence
of obstacles, their distances, their relative size based on distance,
their speed, and so on. A human being also recognizes
what he or she is seeing, So if a human
being sees a fence, you can accurately predict with high
(03:23):
certainty that that fence is not going to jump out
into the middle of the road and do something to you.
On the other hand, if there's a child playing with
a ball on a sidewalk, it's a completely different story,
and humans can recognize that instantly. Computer vision systems are
nowhere near this point in their development, so they rely
on extra sensors to provide more information. Self driving cars
(03:46):
do have camera based vision systems, and they use it
to see other cars, unexpected obstacles, road marking signs, and
so on. But in addition, self driving cars almost always
have GPS sense, or so they have a better idea
of exactly where they are and where they're pointing. They
also have light our systems laser scanners that can look
(04:08):
for nearby objects and accurately judge their distance. They also
have radar systems normally, and they may have infrared sensors
to improve night vision. These sensors all feed into a
powerful onboard computer, often multiple computers that are usually located
back in the back catch area of the trunk. They
(04:29):
could fill an entire trunk in fact, and these process
all the data that the sensors are gathering. The internal
computer will have access to a database of maps and
other relevant information as well. For example, engineers may pre
drive or route and pre catalog all the signs, road markings, curves, crosswalks,
(04:50):
traffic lights, and so on that the self driving car
will encounter along the road. This way, the car knows
what to expect and complan accordingly. Finally, the onboard computer
may also be communicating by radio, with bigger computers holding
even more data in nearby buildings. You might wonder why
the driving task, which seems effortless to just about any
(05:14):
normal human adult, would be so hard for a computer.
It's because the human brain is deceptive. Although it's small
and silent compared to a desktop computer, the human brain
still has more power than mankind's biggest fills a whole
building supercomputers. The massive amount of computing power in the
(05:34):
human brain is what makes driving appear to be so
easy for us. In thirty or forty years, our computers
will have that same kind of power in a small
package if things keep progressing at their current rate. Be
sure to check out our new video podcast, Stuff from
the Future. Join how Stuff Work staff as we explore
(05:55):
the most promising and perplexing possibilities of tomorrow. The Housuffork's
iPhone app has arrived. Download it today on iTunes.